Energy Prediction Model Research Articles

Using Kinect body joint detection system to predict energy expenditures during physical activities

The purpose of this study was to explore the potential of Kinect body joint detection to facilitate the calculation of energy expenditure during exergame exercises. Two Kinect-based biomechanical models - mechanical energy (KineticE) and work (WorkE) were employed to estimate the energy expenditure during four Wii™ exergame session. Consequently, two stepwise regression models were developed from nineteen participants’ data and then validated by five holdout participants. The data collected using an accelerometer (r = 0.835, p < 0.001) had the highest correlation as compared to that of the WorkE (r = 0.805, p < 0.001) and KineticE (r = 0.466, p < 0.001) correlations with the reference indirect calorimetry using Quark activity energy expenditure (QuarkAEE). The regression results show that KineticE and the weight of the participant were significant factors for mechanical energy prediction (AEEKinetic). However, according to the work prediction equation (AEEWork), only WorkE was significant. The new energy prediction models showed significant agreement with the standard QuarkAEE (AEEKinect, r = 0.641, p = 0.02; AEEWork, r = 0.793, p < 0.001), and they were comparable to accelerometer predictions (r = 0.682, p = 0.001). The findings indicate that Kinect can be a potentially viable alternative to measure energy expenditures. The models can be applied with higher accuracy, especially when the activity demands high body movements.

Active Learning Training Strategy for Predicting O Adsorption Free Energy on Perovskite Catalysts using Inexpensive Catalyst Features

AbstractMachine learning (ML) based energy prediction models are among the most effective descriptor‐based catalyst screening tools for heterogeneous reaction systems. However, their implementations are limited due to expensive data labelling, ab initio feature evaluation and lack of universal catalyst features, that is, beyond d‐band theory. Herein, we propose an inexpensive geometric feature for application on systems beyond d‐band theory, for example perovskites comprising of s‐, p‐, d‐ and f‐block elements. We outline a workflow that inputs these features into an active learning algorithm that enables effective data labelling, whilst improving prediction accuracies of existing models. We then use batch sampling to define termination criteria and to implement time‐series error forecasting for further reducing the number of expensive data labelling for training. We implement this workflow to train ML models for predicting oxygen adsorption free energy on perovskites and achieve similar, if not better, prediction accuracies as obtained from ab initio features.

Urban building energy prediction at neighborhood scale

Urban building energy model has been the focus of much research in recent years, especially using data-driven techniques, however, the success of which needs to solve the recognized challenges, such as sufficient energy use dataset in spatial and temporal scales and mutual effect between inter-buildings. Using monthly and yearly energy data from 539 residential buildings and 153 public buildings in a county-level city, this study investigated five typical data-driven urban building energy prediction models on the neighborhood scale. The k nearest neighbors (KNN), support vector regression (SVR), and long short-term memory (LSTM) algorithms were selected as data-driven predictive techniques. The Model 1 was a data-driven energy prediction model for individual building, and with LSTM, the best results for Model 1 can be averagely 0.41 of MAPE and averagely 0.57 of R2. The Model 2 applied different percentages (100%, 95%, 90%, 85%, 80%, 75%, 70%, 65%, and 60%) of original energy dataset to predict total energy demand. The results for Model 2 are averagely 0.065 of MAPE and averagely 0.95 of R2, which also proved that reducing the size of dataset did not influence the results. The Model 3 and 4 created building networks with energy data and building morphology, respectively, and integrated them in urban building energy prediction models, The MAPE results are mostly lower than 0.4 and 0.36, respectively, and R2 results are mostly higher than 0.85 and 0.8 for Model 3 and 4, respectively. The Model 5 combined building morphological metrics and yearly energy data, which received 0.093 and 0.194 of MAPE results and 0.975 and 0.99 of R2 for residential and public buildings, respectively. Finally, this study can contribute to provide more solutions to urban building energy prediction while reduce the high data requirements of urban energy models.

Bi-Level Poisoning Attack Model and Countermeasure for Appliance Consumption Data of Smart Homes

Accurate building energy prediction is useful in various applications starting from building energy automation and management to optimal storage control. However, vulnerabilities should be considered when designing building energy prediction models, as intelligent attackers can deliberately influence the model performance using sophisticated attack models. These may consequently degrade the prediction accuracy, which may affect the efficiency and performance of the building energy management systems. In this paper, we investigate the impact of bi-level poisoning attacks on regression models of energy usage obtained from household appliances. Furthermore, an effective countermeasure against the poisoning attacks on the prediction model is proposed in this paper. Attacks and defenses are evaluated on a benchmark dataset. Experimental results show that an intelligent cyber-attacker can poison the prediction model to manipulate the decision. However, our proposed solution successfully ensures defense against such poisoning attacks effectively compared to other benchmark techniques.

Attention-based interpretable neural network for building cooling load prediction

Machine learning has gained increasing popularity in building energy management due to its powerful capability and flexibility in model development as well as the rich data available in modern buildings. While machine learning is becoming more powerful, the models developed, especially artificial neural networks like Recurrent Neural Networks (RNN), are becoming more complex, resulting in “darker models” with lower model interpretability. The sophisticated inference mechanism behind machine learning prevents ordinary building professionals from understanding the models, thereby lowering trust in the predictions made. To address this, attention mechanisms have been widely implemented to improve the interpretability of deep learning; these mechanisms enable a deep learning-based model to track how different inputs influence outputs at each step of inference.This paper proposes a novel neural network architecture with an attention mechanism for developing RNN-based building energy prediction, and investigates the effectiveness of this attention mechanism in improving the interpretability of RNN models developed for 24-hour ahead building cooling load prediction. To better understand, explain and evaluate these neural network-based building energy prediction models, the obtained attention vectors (or metric) are used to visualize the influence of different parts of model inputs on the prediction result. This helps the users to understand why predictions are made by the model, as well as how input sequences proportionally influence the output sequences. Further analysis of attention vectors can provide interesting temporal information for understanding building thermal dynamics, like the thermal inertia of the building. The proposed attention-based architecture can be implemented in developing optimal operation control strategies and improving demand and supply management. The model developed based on this architecture is assessed using real building operational data, and shows improved accuracy and interpretability over baseline models (without adopting attention mechanisms). The research results help to bridge the gap between building professionals and advanced machine learning techniques. The insights obtained can be used as guidance for the development, fine-tuning, explanation and debugging of data-driven building energy prediction models.

상호운용적 스마트 공장을 위한 PMML·PFA와 자산관리쉘 통합 방법

The present work proposes a method to integrate Predictive Model Markup Language (PMML) and Portable Format for Analytics (PFA) with Asset Administration Shell (AAS) for interoperable smart factories. This method enables the exchange and sharing of data analytics models across heterogeneous manufacturing assets through their deployment into AAS, which specifies a unified and standardized format to represent digital models, data and technical functionalities of physical assets. This article includes : the conceptual model and process, information structure and system architecture for creating and deploying PMML and PFA-based data analytics models into AAS. This article also includes a prototype implementation. In the prototype, a server creates regression and neural network-driven energy prediction models for a production machine, represents the models based on the PMML and PFA schema, generates and responds to an AAS instance with its submodels embedding such PMML and PFA models; meanwhile, a client requests and receives the AAS instance on a web environment.

Energy Prediction Model of PSO-BP Neural Network Three-dimensional Clusters based on Atomic Coordinates

Energy prediction for different cluster structures is the basis for finding and predicting the global optimal structure of clusters. The current methods for predicting the energy of the ground state structures of different clusters include theoretical prediction methods and optimized simplified potential energy function methods. The accuracy of the theoretical prediction method is high, but its calculation amount is too large. Therefore, this paper proposes a PSO-BP neural network three-dimensional cluster energy prediction model based on atomic coordinates, and uses different types of Euclidean distances between atoms as input variables, and the energy of clusters with different structures as output variables. Select gold cluster Au20 and boron cluster B45-part of the sample data as the training set to build the model, and predict the rest of the samples, and finally get: the prediction accuracy of the PSO-BP neural network model is higher than that of the traditional BP neural network model. The cluster energy prediction model is feasible.

Prediction of the Charpy V-notch impact energy of low carbon steel using a shallow neural network and deep learning

The impact energy prediction model of low carbon steel was investigated based on industrial data. A three-layer neural network, extreme learning machine, and deep neural network were compared with different activation functions, structure parameters, and training functions. Bayesian optimization was used to determine the optimal hyper-parameters of the deep neural network. The model with the best performance was applied to investigate the importance of process parameter variables on the impact energy of low carbon steel. The results show that the deep neural network obtains better prediction results than those of a shallow neural network because of the multiple hidden layers improving the learning ability of the model. Among the models, the Bayesian optimization deep neural network achieves the highest correlation coefficient of 0.9536, the lowest mean absolute relative error of 0.0843, and the lowest root mean square error of 17.34 J for predicting the impact energy of low carbon steel. Among the variables, the main factors affecting the impact energy of low carbon steel with a final thickness of 7.5 mm are the thickness of the original slab, the thickness of intermediate slab, and the rough rolling exit temperature from the specific hot rolling production line.

Assessment of Computational Tools for Predicting Supramolecular Synthons

The ability to predict the most likely supramolecular synthons in a crystalline solid is a valuable starting point for subsequently predicting the full crystal structure of a molecule with multiple competing molecular recognition sites. Energy and informatics-based prediction models based on molecular electrostatic potentials (MEPs), hydrogen-bond energies (HBE), hydrogen-bond propensity (HBP), and hydrogen-bond coordination (HBC) were applied to the crystal structures of twelve pyrazole-based molecules. HBE, the most successful method, correctly predicted 100% of the experimentally observed primary intermolecular-interactions, followed by HBP (87.5%), and HBC = MEPs (62.5%). A further HBC analysis suggested a risk of synthon crossover and synthon polymorphism in molecules with multiple binding sites. These easy-to-use models (based on just 2-D chemical structure) can offer a valuable risk assessment of potential formulation challenges.

Peeking inside the black-box: Explainable machine learning applied to household transportation energy consumption

Sustainability policies to mitigate transportation energy impacts on the urban environment are urgently needed. Energy prediction models provide critical information to decision-makers who develop sustainability policies to reduce energy use and emissions. We present a transportation energy model (TEM) that uses Explainable Artificial Intelligence (XAI) methods to predict household transportation energy consumption in this study. The TEM model uses data-driven approaches for household transportation energy prediction. Machine learning techniques in artificial intelligence (AI) predictive modeling have become popular due to their ability to capture nonlinear and complex relationships. On the other hand, developing comprehensive understanding the inference mechanisms in AI models and ensuring trust in their predictions is challenging. This is because AI models are mostly of high complexity and low interpretability; in other words, they are black-box models. This study presents a case study of how model transparency and explanation can be generated using the Local Interpretable Model-Agnostic Explanation (LIME) to support advanced machine learning techniques in the transportation energy field. The methodology has been implemented based on the Household Travel Survey (HTS) data, which is used to train the artificial neural network with a relatively high degree of accuracy. The importance and effect (local explanation) of HTS inputs (such as household travel, demographics, and neighborhood data) on transportation energy consumption for specific traffic analysis zones (TAZs) are analyzed. The results are valuable to promote intelligent and user-friendly transportation energy planning models in urban regions across the world.

An energy and time prediction model for remanufacturing process using graphical evaluation and review technique (GERT) with multivariant uncertainties.

The rising energy price and stringent energy efficiency-related legislations encourage decision makers to concern more about energy efficiency in current manufacturing competition. In this regard, a quick and accurate prediction of the energy consumption and makespan in the manufacturing process has been a prerequisite for energy optimization. Given the various types of uncertainties in the remanufacturing system such as stochastic, fuzzy, and grey factors, the present study developed a prediction model that forecasts the energy consumption, completion time, and probability of processing routes. It adopted the graphical evaluation and review technique (GERT) to convert remanufacturing process into an uncertain network, considering multivariant uncertainties instead of merely stochastic uncertainty in prior works. We provided a generic seven steps to implement this approach. The energy consumption and completion time of remanufacturing process were determined in conjunction with Mason's rule and chance-constrained programming. Connecting rod reprocessing was presented as a numerical example. Based on the GERT network, we conducted an Arena simulation to validate the feasibility and effectiveness of this approach. In addition, we adopted the concept of criticality index to conduct sensitivity analysis and examine the predominant factors affecting the concerned indicators. This study would enable remanufacturers to perform a quick prediction of energy use and makespan in remanufacturing process.

Evaluation and Prediction of Energy Content of Municipal Solid Waste: A review

Researches in the literature have unveiled the potential of resource and energy recovery from waste, it can therefore no longer be regarded as trash. This study reviews the literature to evaluate and analyses studies which estimated the experimental heating value of waste and the theoretical energy potential recoverable from waste through thermochemical and biochemical routes at different case studies. It was observed in this study that most developing countries are not exploiting the full potential of energy recoverable from waste. Models developed to predict the energy content of municipal solid waste (MSW) based on the elemental analysis, proximate analysis and physical composition were evaluated. A comparative analysis of the energy prediction models was also done. Artificial neural network (ANN) and multiple linear regressions found more applications in energy prediction. Energy prediction based on ultimate analysis using the elemental composition of the waste was predominant and are the most accurate; while proximate analysis based predictions were the least. The prediction accuracy of ANN is greater than the linear regression in the forecast of the energy content of MSW. However, a major limitation in the use of these modelling techniques was identified. Most of the generalized models may not capture the peculiarity of the waste generated at a particular place or municipality and therefore may not be very accurate for specific applications at some municipalities.

Contribution of households' occupant profile in predictions of energy consumption in residential buildings: A statistical approach from Mediterranean survey data

There is a growing acceptance that the profile of occupants has a significant influence on the variation in energy consumption in buildings. This is even more evident in the housing sector, where the characteristics of the users of dwellings differ widely from one profile of inhabitant to another, due not only to varied lifestyles, but also to reasons related to geography, culture, behaviour and socio-economic status. This study addresses the socio-demographic factors of occupants that explain the variability in the energy performance of residential buildings and seeks to answer two main research questions: Q1 What kind of variables have the most influence on the household’s energy consumption? Q2 Does the inclusion of occupant profile-related variables improve the energy consumption prediction model? To answer both questions, the study proposed a methodology to develop a set of energy prediction models based on the statistical analysis of real energy consumption data. The method was subsequently applied to the context of the Mediterranean region and serves as an effective tool for quantitatively assessing the influence of occupant characteristics in real building energy consumption. It also offers valuable background information to help practitioners make decisions in the energy policy and building energy performance contexts.

Dataset Construction to Explore Chemical Space with 3D Geometry and Deep Learning.

A dataset is the basis of deep learning model development, and the success of deep learning models heavily relies on the quality and size of the dataset. In this work, we present a new data preparation protocol and build a large fragment-based dataset Frag20, which consists of optimized 3D geometries and calculated molecular properties from Merck molecular force field (MMFF) and DFT at the B3LYP/6-31G* level of theory for more than half a million molecules composed of H, B, C, O, N, F, P, S, Cl, and Br with no larger than 20 heavy atoms. Based on the new dataset, we develop robust molecular energy prediction models using a simplified PhysNet architecture for both DFT-optimized and MMFF-optimized geometries, which achieve better than or close to chemical accuracy (1 kcal/mol) on multiple test sets, including CSD20 and Plati20 based on experimental crystal structures.

A nanosatellite task scheduling framework to improve mission value using fuzzy constraints

Task scheduling is an effective approach to increase the value of a satellite mission, which leads to improved resource management and quality of service. This work improves the energy prediction model and a task scheduling formulation, expressed in integer programming to maximize the number of tasks performed in nanosatellite missions. A realistic battery model is introduced in the formulation to extend battery lifetime. This is achieved by a disjunctive program that makes battery charge and discharge more efficient. Furthermore, fuzzy constraints are designed to limit the current rates (for charge and discharge) and the depth of discharge for battery lifetime preservation. Each battery access is penalized in the objective function, thereby stimulating energy consumption to match energy input. For simulation purposes, the varying power input was based on two-line element data of the CubeSat FloripaSat-I, operating in an orbit with J2 perturbation and an attitude that keeps one face of the nanosatellite towards the Earth for the entire orbit, similar to a remote sensing mission. The effectiveness of the task-scheduling methodology was shown by means of simulated experiments of representative scenarios. With the improvements proposed here, a robust and realistic framework for optimal offline scheduling of nanosatellite missions is achieved.

Ensemble Prediction Approach Based on Learning to Statistical Model for Efficient Building Energy Consumption Management

With the development of modern power systems (smart grid), energy consumption prediction becomes an essential aspect of resource planning and operations. In the last few decades, industrial and commercial buildings have thoroughly been investigated for consumption patterns. However, due to the unavailability of data, the residential buildings could not get much attention. During the last few years, many solutions have been devised for predicting electric consumption; however, it remains a challenging task due to the dynamic nature of residential consumption patterns. Therefore, a more robust solution is required to improve the model performance and achieve a better prediction accuracy. This paper presents an ensemble approach based on learning to a statistical model to predict the short-term energy consumption of a multifamily residential building. Our proposed approach utilizes Long Short-Term Memory (LSTM) and Kalman Filter (KF) to build an ensemble prediction model to predict short term energy demands of multifamily residential buildings. The proposed approach uses real energy data acquired from the multifamily residential building, South Korea. Different statistical measures are used, such as mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and R2 score, to evaluate the performance of the proposed approach and compare it with existing models. The experimental results reveal that the proposed approach predicts accurately and outperforms the existing models. Furthermore, a comparative analysis is performed to evaluate and compare the proposed model with conventional machine learning models. The experimental results show the effectiveness and significance of the proposed approach compared to existing energy prediction models. The proposed approach will support energy management to effectively plan and manage the energy supply and demands of multifamily residential buildings.

Development of a multi-granularity energy forecasting toolkit for demand response baseline calculation

The peak load caused by heating, ventilation, and air-conditioning (HVAC) systems is one of the main control targets of a demand response (DR) program. One key issue related to DR is the baseline energy consumption forecasting based on which the DR strategies and performance can be evaluated. Data-driven models, as a promising method for HVAC energy prediction, have been widely studied. But most existing researches have focused on developing complicated algorithms rather than exploring informative features. In this study, a comprehensive review of feature engineering for HVAC energy prediction model development is presented. A novel feature engineering method is roposed. Besides, an easy-to-use, high-accuracy HVAC energy forecasting toolkit that is applicable to datasets of various granularities is developed. This toolkit uses easily available meteorological parameters and raw historical energy data as inputs, on which it performs data preprocessing, feature extension, and integrated optimization, thereby producing the predicted data. By employing a novel feature extension strategy and integrated optimization of feature selection and hyperparameter tuning, this toolkit performs capably in terms of prediction accuracy and stability. The results of a comparative experiment conducted on large-scale data verify that the average forecasting error (measured in terms of the coefficient of variation of the root mean square error) is <8%.

Research on vibration effect of tunnel blasting based on an improved Hilbert–Huang transform

Through a tunnel-blasting project, the effect of tunnel-blasting vibration has been analyzed from the perspective of vibration energy transfer. The non-linear regression method was used to obtain the prediction equation for blast vibration velocity based on the field blast vibration data. Then, the maximum charge per delay of the blasting construction of the tunnel was obtained through the formula inversion. Based on the traditional Hilbert transform, a novel Hilbert–Huang transform (HHT) analysis method considering Complete Ensemble Empirical Mode Decomposition with Adaptive Noise decomposition (CEEMDAN) and wavelet packet threshold de-noising method has been proposed, the feasibility of which was verified using the field blast vibration signals. It has been proved that the improved HHT analysis method can be used to analyze the influence of the different blasting vibration parameters on the regular distribution of vibration energy. In addition, the dimensional analysis method was used to establish the blasting vibration energy prediction model. The results of this research show that the improved HHT analysis method can solve the problem of modal aliasing caused by the traditional decomposition method, and can obtain the purified main modal components, which improves the adaptability of HHT analysis. In addition, with the increase of the distance between blast area and the maximum charge per delay, the high-frequency energy of the blasting signal gradually weakens, while the dominant energy frequency band diverts to the low-frequency band. The methods and conclusions of this research can provide a certain reference for the controlled blasting construction in similar cases.

Quantitative evaluation of the building energy performance based on short-term energy predictions

Building energy prediction is a potential tool for benchmarking the future energy uses of individual buildings. However, inevitable gaps between predicted and actual energy uses and discrepancies between buildings make it challenging to quantify energy consumption. Therefore, this paper proposes a systematic methodology of quantitative building energy evaluation based on short-term energy prediction. First, the 24-h ahead building energy prediction model is developed based on a recurrent neural network via the Multi-Input Multi-Output strategy. Second, the quantitative energy evaluation strategy is proposed to quantify prediction gaps based on the 1-D k-means clustering. Third, case studies are conducted on five real buildings to verify the reliability of the proposed methodology. Results show that the energy prediction models achieved outstanding accuracies. Besides, it is necessary to analyze the absolute percentage error (APE) variation of each time step to deeply understand the building energy performance rather than the overall prediction performance evaluation index, such as CV-RMSE and MAPE. Further, customized energy quantification systems are established for buildings per their specific, individual energy performance. The building energy is quantified by labeling APEs into multiple levels. Moreover, building operation characteristics can be further understood by quantifying energy uses.

Abrasiveness evaluation of complex strata based on PDC special energy consumption

The rock abrasiveness is determined with various methods, which results in no uniform standard formed. In order to predict the PDC bit wear and the abrasiveness of complex strata, the force analysis of the PDC broken rock was carried out, and the bit axial load and torque were obtained, respectively. Then, the calculation model of PDC special energy consumption is deduced. Combining the typical distribution patterns of different strata, the specific energy prediction model of complex strata is established. The wear coefficient of PDC is obtained by the relationship between the triaxial compressive strength of rock and the specific energy of rock breaking, and the grading coefficient and grading standard of abrasiveness of the composite formation are derived, which led to the grading coefficient and grading standard of complex strata abrasiveness. The field test of a well in the north of Songliao Basin verified the rationality and effectiveness of the abrasiveness grading standards and provided the basis for the selection and design of the bit.